ML-systems researcher: asynchronous self-speculative decoding for long-context LLM inference
1
papers here
1
reviewed by Evolve
1
first-author
0
corresponding
—
works · OpenAlex
—
h-index · OpenAlex
Bibliometric figures come from OpenAlex (2026-09-20) and cover the author's entire public record, not just the papers listed here. Reproducibility signal comes from Evolve's own reviews.
Research profile
Output—
Not yet scored: beyond the single reviewed paper, the bibliometric snapshot contains no publication-volume data for this researcher. This axis would measure total research output over time — paper counts, venue quality, and first- or corresponding-author share.
Impact—
Not yet scored: no citation or influence data exists in the current snapshot. This axis would measure how widely the work is cited and adopted — citation counts, citation velocity, and uptake by later inference-systems research.
Focus100/100
Scores 100/100: every cataloged paper sits in a single field — machine learning systems / LLM inference (the field distribution shows one field with full weight). ASPIRE concentrates on speculative decoding for long-context serving, spanning sparse attention, KV cache costs, and batched, per-request scheduling. Such single-field depth is typical of an early-stage profile built around one strong systems thread.
Collaboration36/100
Scores 36/100, built on 5 distinct coauthors in the snapshot. The ASPIRE preprint was written with Hossein Entezari Zarch, Lei Gao, Murali Annavaram, and Salman Avestimehr. A compact, single-paper network yields a modest score — it reflects an early collaboration footprint rather than demonstrated isolation, and should rise if later papers widen the author pool.
Activity—
Not yet scored: the snapshot provides only one publication year (2026), so publishing cadence cannot be computed. This axis would measure research activity over time — outputs per year, continuity, and recent momentum.
Reproducibility100/100
Scores 100/100: a reproducibility mean of 100 computed over 1 paper. Evolve rated ASPIRE's reproducibility readiness as high, judging the reported 1.70–4.58× speedups solid while flagging thin variance reporting. A strong signal, but resting on a single data point rather than an established track record.
Research summary
Amir Ziashahabi is an early-career researcher in machine learning systems, focused on LLM inference efficiency. This profile is name-resolved from public bibliographic data (ORCID 0009-0000-1154-0078); no institutional affiliation appears in the current snapshot. The reviewed record centers on ASPIRE (2026), a first-author arXiv preprint that de-synchronizes batched self-speculative decoding through mixed forwards and per-request scheduling, targeting long-context inference where sparse attention and KV cache management dominate cost. Evolve's review judged the reported 1.70–4.58× speedups as solid while flagging thin variance reporting, and rated reproducibility readiness high — consistent with a reproducibility mean of 100 across the one paper assessed. The radar shows maximal focus (100): all cataloged output falls in a single field, machine learning systems / LLM inference. The collaboration score is 36, built on five distinct coauthors including Hossein Entezari Zarch, Lei Gao, Murali Annavaram, and Salman Avestimehr. Output, impact, and activity axes remain unscored because citation counts, publication volume, and year-by-year productivity are absent from the data. Overall, this is a focused, early-stage systems profile whose evidence base should widen as more papers enter the catalog.
Collaboration pattern
The snapshot records 5 distinct coauthors, all tied to the single reviewed paper. The ASPIRE author list includes Hossein Entezari Zarch, Lei Gao, Murali Annavaram, and Salman Avestimehr. A collaboration score of 36 reflects this compact, single-cluster network — one project, one field — while cross-institution reach stays unmeasured because affiliations are absent from the current data.
Research keywords
Highlights
ASPIRE: de-synchronizing batched self-speculative decoding
Long-context systems focus: sparse attention and KV cache
High reproducibility readiness on the first reviewed paper
Timeline
- 2026Published ASPIRE as first author on arXiv: asynchronous batched self-speculative decoding for long-context LLM inference.
- 2026Evolve review: the 1.70–4.58× speedups were judged solid, with variance reporting flagged as thin.
- 2026Reproducibility readiness rated high (mean 100 over 1 paper); the bibliometric snapshot records 5 distinct coauthors and a single research field.
Papers in the Evolve catalog
Frequent co-authors
About this profile
This profile is AI-structured by Evolve.Pub from public bibliographic data (OpenAlex) and Evolve's own paper reviews. It is not written or endorsed by the researcher unless the profile is claimed. Radar dimensions are computed from real bibliometrics with fixed formulas — raw figures are shown alongside; missing dimensions mean insufficient data, never zero. Errors? Claim the profile to correct or hide it.
- Bibliometric snapshot covers one reviewed paper; radar scores are computed from that limited sample, not a full publication record.
- Output, impact, and activity axes are unscored — citation counts, publication volume, and year-by-year productivity are absent from the data.
- Institution and country are unavailable; the profile is name-resolved from public bibliographic data via ORCID 0009-0000-1154-0078.
- Reproducibility score of 100 derives from a single paper (reproMean 100, reproPapers 1) and should not be read as an established track record.